{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,8,18]],"date-time":"2026-08-18T17:30:49Z","timestamp":1787074249213,"version":"build-2736575974"},"reference-count":30,"publisher":"Public Library of Science (PLoS)","issue":"8","license":[{"start":{"date-parts":[[2026,8,18]],"date-time":"2026-08-18T00:00:00Z","timestamp":1787011200000},"content-version":"vor","delay-in-days":0,"URL":"http:\/\/creativecommons.org\/licenses\/by\/4.0\/"}],"funder":[{"DOI":"10.13039\/501100008951","name":"Werner H. Spross-Stiftung","doi-asserted-by":"publisher","award":["n\/a"],"award-info":[{"award-number":["n\/a"]}],"id":[{"id":"10.13039\/501100008951","id-type":"DOI","asserted-by":"publisher"}]}],"content-domain":{"domain":["www.plosone.org"],"crossmark-restriction":false},"short-container-title":["PLoS One"],"abstract":"<jats:p>Large domain-specific foundation models have been widely adopted for retinal image analysis, yet systematic evidence for their advantage over compact general-purpose architectures remains scarce. We benchmarked nine model configurations spanning 22.8M to 303M parameters (vision transformers, hierarchical Swin Transformers, ConvNeXt, and the domain-specific RETFound models) across four tasks: OCT 8-class disease classification, and three fundus photography tasks (DME severity, glaucoma detection, and DR severity grading). All models were evaluated under identical training conditions, with both pretrained (on natural-domain image datasets) and from-scratch initializations compared using Mann-Whitney U tests. Pretraining improved accuracy by 5.18\u201318.41 percentage points across all tasks (p\u2009&lt;\u20090.05 throughout), with larger benefits for CFP modalities and harder tasks. Compact hierarchical models (27\u201329M parameters) matched or exceeded larger architectures on three of four tasks. For instance, the SwinV2-tiny architecture ranked first on OCT, DME, and GL classification. The domain-specific RETFound model (303M) achieved the highest accuracy only on the most challenging task (DR severity grading, where the most severe class is underrepresented at 8% of images), where it outperformed the best compact model by 1.54 percentage points. These results indicate that compact general-purpose models may be sufficient for most retinal classification benchmarks, and that domain-specific foundation models may add higher value mainly for severity grading tasks with skewed class distributions.<\/jats:p>","DOI":"10.1371\/journal.pone.0356202","type":"journal-article","created":{"date-parts":[[2026,8,18]],"date-time":"2026-08-18T17:21:44Z","timestamp":1787073704000},"page":"e0356202","update-policy":"https:\/\/doi.org\/10.1371\/journal.pone.corrections_policy","source":"Crossref","is-referenced-by-count":0,"title":["Compact vision models match domain-specific foundation models for several retinal imaging classification tasks: A systematic benchmark"],"prefix":"10.1371","volume":"21","author":[{"given":"D\u00e1vid","family":"Isztl","sequence":"first","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0009-0005-3639-8289","authenticated-orcid":true,"given":"Tahm","family":"Spitznagel","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"G\u00e1bor M\u00e1rk","family":"Somfai","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0001-6003-9079","authenticated-orcid":true,"given":"Rui","family":"Santos","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"340","published-online":{"date-parts":[[2026,8,18]]},"reference":[{"key":"pone.0356202.ref001","doi-asserted-by":"crossref","first-page":"109119","DOI":"10.1016\/j.diabres.2021.109119","article-title":"IDF diabetes atlas: global, regional and country-level diabetes prevalence estimates for 2021 and projections for 2045","volume":"183","author":"H Sun","year":"2022","journal-title":"Diabetes Res Clin Pract"},{"issue":"11","key":"pone.0356202.ref002","doi-asserted-by":"crossref","first-page":"2081","DOI":"10.1016\/j.ophtha.2014.05.013","article-title":"Global prevalence of glaucoma and projections of glaucoma burden through 2040: a systematic review and meta-analysis","volume":"121","author":"Y-C Tham","year":"2014","journal-title":"Ophthalmology"},{"issue":"2","key":"pone.0356202.ref003","doi-asserted-by":"crossref","first-page":"6","DOI":"10.1167\/tvst.9.2.6","article-title":"Insights into systemic disease through retinal imaging-based oculomics","volume":"9","author":"SK Wagner","year":"2020","journal-title":"Transl Vis Sci Technol"},{"issue":"6","key":"pone.0356202.ref004","doi-asserted-by":"crossref","first-page":"1427","DOI":"10.1007\/s40123-024-00942-x","article-title":"Oculomics: a crusade against the four horsemen of chronic disease","volume":"13","author":"EJ Patterson","year":"2024","journal-title":"Ophthalmol Ther"},{"key":"pone.0356202.ref005","doi-asserted-by":"crossref","first-page":"101350","DOI":"10.1016\/j.preteyeres.2025.101350","article-title":"Oculomics: current concepts and evidence","volume":"106","author":"Z Zhu","year":"2025","journal-title":"Prog Retin Eye Res"},{"key":"pone.0356202.ref006","doi-asserted-by":"crossref","unstructured":"He K, Zhang X, Ren S, Sun J. Deep residual learning for image recognition. In: 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2016. 770\u20138. https:\/\/doi.org\/10.1109\/cvpr.2016.90","DOI":"10.1109\/CVPR.2016.90"},{"issue":"7981","key":"pone.0356202.ref007","doi-asserted-by":"crossref","first-page":"156","DOI":"10.1038\/s41586-023-06555-x","article-title":"A foundation model for generalizable disease detection from retinal images","volume":"622","author":"Y Zhou","year":"2023","journal-title":"Nature"},{"key":"pone.0356202.ref008","author":"Y Zhou","year":"2025"},{"key":"pone.0356202.ref009","author":"MRH Taher","year":"2021"},{"issue":"2","key":"pone.0356202.ref010","doi-asserted-by":"crossref","first-page":"16","DOI":"10.1167\/tvst.13.2.16","article-title":"Deep learning and machine learning algorithms for retinal image analysis in neurodegenerative disease: systematic review of datasets and models","volume":"13","author":"T Bahr","year":"2024","journal-title":"Transl Vis Sci Technol"},{"key":"pone.0356202.ref011","doi-asserted-by":"crossref","unstructured":"Huang SC, Jensen M, Yeung-Levy S, Lungren MP, Poon H, Chaudhari AS. A systematic review and implementation guidelines of multimodal foundation models in medical imaging. Research Square. 2025.","DOI":"10.21203\/rs.3.rs-5537908\/v1"},{"key":"pone.0356202.ref012","doi-asserted-by":"crossref","unstructured":"Deng J, Dong W, Socher R, Li LJ, Li K, Fei-Fei L. Imagenet: a large-scale hierarchical image database. In: 2009 IEEE conference on computer vision and pattern recognition, 2009. 248\u201355.","DOI":"10.1109\/CVPR.2009.5206848"},{"issue":"1","key":"pone.0356202.ref013","doi-asserted-by":"crossref","first-page":"3637","DOI":"10.1038\/s41598-023-30853-z","article-title":"An interpretable transformer network for the retinal disease classification using optical coherence tomography","volume":"13","author":"J He","year":"2023","journal-title":"Sci Rep"},{"issue":"3","key":"pone.0356202.ref014","doi-asserted-by":"crossref","first-page":"25","DOI":"10.3390\/data3030025","article-title":"Indian Diabetic Retinopathy Image Dataset (IDRiD): a database for diabetic retinopathy screening research","volume":"3","author":"P Porwal","year":"2018","journal-title":"Data"},{"issue":"3","key":"pone.0356202.ref015","doi-asserted-by":"crossref","first-page":"231","DOI":"10.5566\/ias.1155","article-title":"Feedback on a publicly distributed image database: the messidor database","volume":"33","author":"E Decenci\u00e8re","year":"2014","journal-title":"Image Anal Stereol"},{"key":"pone.0356202.ref016","unstructured":"de Vente C, Vermeer KA, Jaccard N, Wang H, Sun H, Khader F. AIROGS: artificial intelligence for robust glaucoma screening challenge. In: 2023. https:\/\/arxiv.org\/abs\/2302.01738"},{"issue":"1","key":"pone.0356202.ref017","doi-asserted-by":"crossref","first-page":"291","DOI":"10.1038\/s41597-022-01388-1","article-title":"PAPILA: dataset with fundus images and clinical data of both eyes of the same patient for glaucoma assessment","volume":"9","author":"O Kovalyk","year":"2022","journal-title":"Sci Data"},{"key":"pone.0356202.ref018","volume-title":"EyePACS, APTOS, Messidor diabetic retinopathy dataset","author":"AS Canipek","year":"2023"},{"key":"pone.0356202.ref019","author":"A Dosovitskiy","year":"2020"},{"key":"pone.0356202.ref020","unstructured":"Oquab M, Darcet T, Moutakanni T, Vo H, Szafraniec M, Khalidov V. DINOv2: Learning Robust Visual Features without Supervision. 2024. https:\/\/doi.org\/arXiv:2304.07193"},{"key":"pone.0356202.ref021","doi-asserted-by":"crossref","unstructured":"Liu Z, Lin Y, Cao Y, Hu H, Wei Y, Zhang Z, et al. Swin transformer: hierarchical vision transformer using shifted windows. In: 2021 IEEE\/CVF International Conference on Computer Vision (ICCV), 2021. 9992\u201310002. https:\/\/doi.org\/10.1109\/iccv48922.2021.00986","DOI":"10.1109\/ICCV48922.2021.00986"},{"key":"pone.0356202.ref022","doi-asserted-by":"crossref","unstructured":"Liu Z, Hu H, Lin Y, Yao Z, Xie Z, Wei Y, et al. Swin transformer V2: scaling up capacity and resolution. In: 2022 IEEE\/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2022. 11999\u20132009. https:\/\/doi.org\/10.1109\/cvpr52688.2022.01170","DOI":"10.1109\/CVPR52688.2022.01170"},{"key":"pone.0356202.ref023","doi-asserted-by":"crossref","unstructured":"Liu Z, Mao H, Wu C-Y, Feichtenhofer C, Darrell T, Xie S. A ConvNet for the 2020s. In: 2022 IEEE\/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2022. 11966\u201376. https:\/\/doi.org\/10.1109\/cvpr52688.2022.01167","DOI":"10.1109\/CVPR52688.2022.01167"},{"key":"pone.0356202.ref024","unstructured":"Loshchilov I, Hutter F. Decoupled weight decay regularization. In: 2019. https:\/\/doi.org\/arXiv:1711.05101"},{"key":"pone.0356202.ref025","unstructured":"Steiner A, Kolesnikov A, Zhai X, Wightman R, Uszkoreit J, Beyer L. How to train your ViT? Data, augmentation, and regularization in vision transformers. 2022. https:\/\/doi.org\/arXiv:2106.10270"},{"issue":"1","key":"pone.0356202.ref026","doi-asserted-by":"crossref","first-page":"29","DOI":"10.1148\/radiology.143.1.7063747","article-title":"The meaning and use of the area under a receiver operating characteristic (ROC) curve","volume":"143","author":"JA Hanley","year":"1982","journal-title":"Radiology"},{"issue":"1","key":"pone.0356202.ref027","first-page":"37","article-title":"Evaluation: from precision, recall and f-measure to roc, informedness, markedness and correlation","volume":"2","author":"DM Powers","year":"2011","journal-title":"J Machine Learn Tech"},{"issue":"1","key":"pone.0356202.ref028","doi-asserted-by":"crossref","first-page":"37","DOI":"10.1177\/001316446002000104","article-title":"A Coefficient of agreement for nominal scales","volume":"20","author":"J Cohen","year":"1960","journal-title":"Edu Psychol Measurement"},{"issue":"1","key":"pone.0356202.ref029","doi-asserted-by":"crossref","first-page":"50","DOI":"10.1214\/aoms\/1177730491","article-title":"On a test of whether one of two random variables is stochastically larger than the other","volume":"18","author":"HB Mann","year":"1947","journal-title":"Ann Math Statist"},{"key":"pone.0356202.ref030","volume-title":"Nonlinear multiobjective optimization","author":"K Miettinen","year":"1999"}],"container-title":["PLOS One"],"original-title":[],"language":"en","link":[{"URL":"https:\/\/dx.plos.org\/10.1371\/journal.pone.0356202","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,8,18]],"date-time":"2026-08-18T17:21:49Z","timestamp":1787073709000},"score":1,"resource":{"primary":{"URL":"https:\/\/dx.plos.org\/10.1371\/journal.pone.0356202"}},"subtitle":[],"editor":[{"given":"Jia-Lang","family":"Xu","sequence":"first","affiliation":[],"role":[{"vocabulary":"crossref","role":"editor"}]}],"short-title":[],"issued":{"date-parts":[[2026,8,18]]},"references-count":30,"journal-issue":{"issue":"8","published-online":{"date-parts":[[2026,8,18]]}},"URL":"https:\/\/doi.org\/10.1371\/journal.pone.0356202","relation":{},"ISSN":["1932-6203"],"issn-type":[{"value":"1932-6203","type":"electronic"}],"subject":[],"published":{"date-parts":[[2026,8,18]]}}}